Google will allow the removal of the visible Gemini branding, but will maintain invisible identifiers to detect generated content
The removal of the visible watermark could make it harder to quickly identify an image or video generated with AI. However, Google has internal mechanisms to retain that information.
The company made this decision while other image generators with artificial intelligence also do not use visible watermarks. Among them are OpenAI and Meta, which use different systems to identify AI-generated content.
Google will allow users to disable the visible watermarks on images, videos, and music generated with its artificial intelligence tools.
The company will incorporate a new option in Gemini and Flow to hide the watermark in the shape of a flash that appears on content created with its models.
The update adds an option called “Media watermark”, which allows users to turn the visible watermark on or off. When removed, the label located in the bottom right corner of the generated content disappears.
This feature applies to material created with the Nano Banana and Omni models. Google also plans to bring this setting to Search.
Although the visible watermark will be optional, Google will continue to incorporate identifiers that cannot be seen with the naked eye. Content generated with AI will retain SynthID watermarks and C2PA metadata.
As explained by Josh Woodward, vice president of Google Labs, Gemini, and AI Studio, these mechanisms will allow verification of whether a piece of content was generated using artificial intelligence.
The verification can be done through Gemini or Google Search, checking if the content has any of these identifiers.
OpenAI uses SynthID and C2PA to indicate whether content was generated with AI.
Meta developed its own standard Content Seal to detect and label content.
Anthropic announced the incorporation of invisible watermarks in texts and images generated with AI.
Google plans to implement the watermark setting also in Search. However, the switch will not be available in countries where regulations require that AI-generated content has a visible watermark.
In this way, the company will allow users to remove the label they see on their content, but will keep SynthID and C2PA as background identification mechanisms.
